GPT Pro Combines PSGD-Kron and KL-Shampoo into Decomposition-Free Optimizer

A researcher used GPT Pro to combine PSGD-Kron's Σ^{-1/2} multiplicative updates with KL-Shampoo's KL objective, yielding a new optimizer that requires no eigendecomposition, matrix inversion, or linear system solving.

2026-08-26 ~ 2026-08-26 · 4 related posts